基于领域自适应的风力机发电机轴承故障诊断方法研究  

RESEARCH ON FAULT DIAGNOSIS METHOD OF WIND TURBINE GENERATOR BEARINGS BASED ON DOMAIN ADAPTATION

作　　者：田淼 苏晓明 陈长征[1,2] 安文杰 孙鲜明 Tian Miao;Su Xiaoming;Chen Changzheng;An Wenjie;Sun Xianming(School of Mechanical Engineering,Shenyang University of Technology,Shenyang 110870,China;Liaoning Vibration and Noise Control Engineering Research Center,Shenyang 110870,China;Ningbo Kunbo Measurement and Control Technology Co.,Ltd.,Ningbo 315200,China)

机构地区：[1]沈阳工业大学机械工程学院,沈阳110870 [2]辽宁省振动噪声控制工程技术研究中心,沈阳110870 [3]宁波坤博测控科技有限公司,宁波315200

出　　处：《太阳能学报》2023年第11期310-317,共8页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(51675350,51575361)。

摘　　要：针对不同型号的风力机发电机滚动轴承采集的振动信号存在分布差异且待诊断轴承样本标签不足的问题,提出一种基于聚类领域自适应卷积神经网络(CDA-CNN)的风力机发电机滚动轴承故障诊断方法。首先利用一维卷积神经网络提取源域中有标签轴承数据和目标域中无标签轴承数据的特征,其次利用聚类方法减小数据特征的条件分布差异并为目标域数据提供伪标签,随后利用最大均值差异(MMD)对齐两域的边缘分布,最终得到风力机发电机滚动轴承的故障诊断模型。将所提出的CDA-CNN对实际风力机发电机滚动轴承进行故障诊断,诊断结果表明:所提出方法的故障诊断精度高达92.52%,有效解决了可用数据标签不足的问题。试验对比结果表明:CDA-CNN模型的诊断精度和迁移性均优于其他方法,对风力机发电机滚动轴承的故障诊断研究具有一定的工程应用价值。Aiming at the problem that the vibration signals collected by different types of wind turbine generator rolling bearings have different distribution and the sample labels of rolling bearings to be diagnosed are insufficient,this paper proposed a fault diagnosis method of wind turbine generator rolling bearings based on clustering domain adaptive convolutional neural network(CDA-CNN).Firstly,the features of labeled bearing data in the source domain and unlabeled bearing data in the target domain were extracted by using the 1D convolutional neural network.Secondly,the clustering method was used to reduce the difference in the conditional distribution of data features and provided pseudo labels for target domain data.Then,the maximum mean difference(MMD)was used to align the edge distribution of the two domains.Finally,the fault diagnosis model of wind turbine generator rolling bearings was obtained.The proposed CDA-CNN is applied to the fault diagnosis of actual wind turbine generator rolling bearings.The diagnosis results show that the fault diagnosis accuracy of the proposed method is as high as 92.52%,which effectively solves the problem of insufficient available data labels.The test results show that the diagnostic accuracy and transfer of CDA-CNN are better than other methods,and it has a certain engineering application value for the fault diagnosis of wind turbine generator rolling bearings.

关 键 词：风力机 滚动轴承 故障诊断 领域自适应 聚类 

分 类 号：TH133.33[机械工程—机械制造及自动化] TH17